ChatGPT and co. haven't settled on holiday hotels yet

Key findings

• Ask twice, get two answers. Ask the same assistant the same question twice: at the median, the hotels and restaurants named overlap by only about one fifth.
• Each assistant names different places. For the same question on the same day, ChatGPT and Claude have no business in common in 44 percent of cases.
• No fixed regional favourite. For no hotel topic is the gap between provinces larger than the margin of error. For winter holiday destinations, by contrast, about half of the places named are in Tyrol.
• The assistants look in different places. ChatGPT cites the businesses' own websites 55 percent of the time, Claude relies more on booking portals and travel guides.
• Some hotels still stand out. For questions about hotels in Tyrol, Interalpen-Hotel Tyrol appears in 45 of 198 answers.

Innsbruck, 28 September 2026. Ask ChatGPT, Claude or Gemini about a holiday hotel in Austria and the names keep changing. That is the result of a data study by the PR analytics platform aclipp, carried out around the FAFGA tourism trade fair in Innsbruck. From 17 to 20 September 2026, aclipp asked the three assistants typical traveller questions about holiday hotels, destinations and restaurants in Austria, each question several times, and analysed 2,952 answers. Ask the same question twice and, at the median, only about one in five businesses named shows up again, counting hotels and restaurants together.

Ask twice, get two answers

Ask the same assistant the same question twice: at the median, the hotels and restaurants named overlap by only about one fifth. In 20 to 25 percent of cases there is no overlap at all.

The lists also differ between assistants. For the same question on the same day, ChatGPT and Claude have no business in common in 44 percent of cases, Claude and Gemini in 50 percent, ChatGPT and Gemini in 34 percent. Being named by one assistant does not mean being named by the next.

"The AI assistants haven't settled on hotels yet," says Sascha Kirstein, founder and CEO of aclipp. "Ask the same question twice and only about one in five hotel and restaurant names comes up again, and each assistant names different places. Nobody in the hotel industry has clearly won AI visibility yet. Whoever starts working on it now finds an open field."

The destination is set, the individual hotel is not

For family hotels in Austria, ChatGPT and Gemini put Salzburg ahead, Claude puts Tyrol ahead. For no hotel topic is the gap between provinces larger than the margin of error. For winter holiday destinations the picture is clear: about half of the places named are in Tyrol. In AI answers, the destination is set, the individual hotel is not.

Some hotels still stand out. For questions about hotels in Tyrol, Interalpen-Hotel Tyrol is named in 45 of 198 answers, well ahead of all others. Behind it, Alpen-Herz (Ladis), Das Central (Sölden) and SPA-Hotel Jagdhof (Neustift) are almost level.

Where the assistants look

ChatGPT cites the businesses' own websites 55 percent of the time. Claude relies more on booking and review portals (38 percent) and editorial travel guides (26 percent). The source figures are approximations and apply to all questions, not only to hotels. Gemini does not return sources through the query used. Which source triggers a mention cannot be read from this.

"More than half the time, ChatGPT reads directly on the businesses' own websites, Claude more on booking portals and in travel guides," Kirstein says. "That makes a hotel's own website the source it has the most influence over."

What hotels can take from this

• Keep your own website complete: location, target groups, offers and facilities in clear language and readable for machines.
• Maintain your listings on booking and review portals, because assistants like Claude read there.
• Get featured in editorial travel guides and trade media.
• Don't judge by a single answer: mentions vary from query to query, so only repeated checks across several assistants are meaningful.

"At FAFGA, hoteliers were very interested in the topic, but many don't know where to start yet," Kirstein adds. "The first step is to know whether and how often your own hotel is named at all. And a single answer is not a ranking: if you're in it, don't celebrate too early, and if you're missing, don't worry too early."

Methodology

• Study: aclipp, own measurement with the aclipp AI Visibility feature, not commissioned by anyone
• Period: 17 to 20 September 2026
• AI assistants: ChatGPT, Claude and Gemini
• Questions: 116 typical traveller questions about hotels, holiday destinations and restaurants, all about holidays in Austria, in German and English, with location settings Germany, Austria and the United Kingdom. 199 combinations, each asked three to six times per assistant.
• Data: 2,952 analysed answers
• Stability: Unlike studies that rate individual hotels with a score, this study also measures how stable the mentions are: the same question several times, across three assistants. The overlap figures refer to the businesses named and only count pairs in which both answers name at least one business.
• Analysis: The names mentioned were extracted with a language model and checked automatically against the original text; the hotel names published were additionally checked via web search. There was no manual review of the extraction, so no error rate is given.
• Questions: compiled by aclipp and modelled on typical traveller questions, but not a sample of real user queries.
• Interpretation: Mentioned doesn't mean best. The study measures which names the assistants give, not the quality of a hotel.

The study's Tyrol results (destinations, ski areas, restaurants) and all charts are in the press release on Tyrol. The data package and all raw data (questions and answers) are available on request. Sascha Kirstein is available for interviews and background conversations.